Re: [PATCH] Revert 9fc2105aeaaf56b0cf75296a84702d0f9e64437b to fix pyaudio (and probably more)

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, Jan 07, 2015 at 04:15:00PM -0500, Nicolas Pitre wrote:
> 
> > However, if running userland on platform A works, and but it doesn't
> > work on platform B.  The breakage may well be due to platform A reporting
> > 300 bogomips because it's using the kernel software loop, and platform
> > B reporting 6 bogomips because its using a hardware timer, but the CPU
> > is actually faster.  However, this is not a kernel problem, and it
> > certainly is not a regression.  It's a userspace bug which needs
> > userspace to fix.
> 
> There I disagree.  In the spirit of "the kernel shall never break user 
> space ever" I'd say that the kernel is simply doing a poor job at 
> providing user space with a value that won't break user space 
> expectations.  And since it is not that hard to do (I made a patch 
> already) I'd say we have less to lose by fixing it than keeping a 
> totally senseless value around.

It's not that the kernel shall never break userspace. It must not cause
userspace regressions.

If application A worked on box X and they upgrade the kernel and then
application A no longer works. That's a regression, and must be fixed.

Now if I understand what Russell stated, if application A works on box X
and you move to box Y using the same kernel, and application A no longer
works, that's not a regression with the kernel (unless it use to work
on box Y). If it never worked on box Y, it's a platform issue and
application A is not robust enough to deal with it. AKA, not a kernel
bug.

Now, it gets interesting if a fix was made that lets application A work
on box Y, but that fix broked application B on box X. Reverting that
fix will cause a regression for A or Y, but the fix itself caused a
regression for B on X. In that case we need a new fix (which may be
the case we are here).

-- Steve